Oh, boy...
There used to be some in-depth threads in the archives, but I think they all got lost when Werner had to switch to the new forum system.
First, read the threads on buying used robots. There are a lot of hidden costs, and things to watch out for.
Second, you are NOT going to get the kind of accuracy even a cheap CNC router would achieve. Robots have repeatability/precision, but not accuracy. Not without a lot of extra work and money. On top of that, robots don't have anything like the rigidity of CNC machines (except maybe the super-cheap Chinesium CNCs). Robot aren't made to resist external forces, like that a milling spindle will put on them. Just changing from Climb to Conventional cutting could change the cut your robot makes by millimeters. At minimum, expect to need a lot of spring passes.
Third, the robot will only be a part of the cost -- the software you need to produce a robot motion program from a 3D model (as opposed to a CNC machine) can easily cost $20k or more, and takes a lot of training and skill to operate. You might be able to use RoboDK, which is the cheapest offline programming system I know of, but there's going to be a lot of learning curve even with that. And unless you buy one of the major robot brands, you might easily find your cheap used robot isn't supported. Or its operating system is too out of date to work with modern program generators.
Fourth (should probably be first), SAFETY. These machines can kill you, a whole lot easier than any kind of CNC -- a CNC can't reach over it's back, punch through a fence like tissue paper, and hit you 2-3m away, a robot can. And all it takes is one little programming mistake (which is even easier with offline program generation). Even for a single-robot cell, making it properly safe can cost nearly as much as the robot costs itself. Fence it off, add hard stops so it can't punch through the fence, add some E-Stops and absolutely safety-wire the gate. For the love of all that's holy, take this seriously. The only robot-related fatalities in the past decade have been people who ignored or bypassed safeties, and got within the robot's reach thinking they knew were it would/wouldn't go, or that they could dodge. They were fatally wrong.
Bottom line: people have done this, but it's very much the Fast/Cheap/Good triangle. Pick two. Maybe more like one and a half. Either you go Full Pro and pay $$$$$$ for all the "right" gear, or you DIY it and spend a lot of blood, sweat, and tears getting something usable through trial and error.
I have a personal bias in favor of KUKAs, so take this with a bit of salt. That said, the 150 and 210 are generally fine machines, and at least as accurate and rigid as any other major-brand robot in the same weight class -- probably a bit more, in fact. But! Used robots can have a lot of wear&tear, often not obvious. A robot that's been in an automotive plant for 5 years probably has 20,000-40,000 hours of runtime, and (given the nature of automotive lines) most of that probably at max speed, max acceleration, "slamming" all over in an attempt to make an unreasonable cycle time. And given how even a clapped-out robot can still give decent repeatable positioning, touching up the points can compensate for the wear until the seals start leaking or the bearings start to seize.
For you, this is a major point, b/c while 90% of normal robot applications rely on repeatability, any CNC-ish application is going to rely on accuracy. And as a robot wears, its accuracy degrades much faster than its repeatability.
The bigger issue, though, is the controller. You do not want an obsolete controller, they're hard to get parts, service, or software updates for. For KUKAbots, don't even think about a KRC1 unless you know you can get experienced help. Even a KRC2 would be iffy, at this point, although I'd be willing to take on a late-model KRC2. Ditto for "special model" units -- Volkswagen, for example, sells lots of used KUKAs, but they're all "special" units that have been modified to VW's unique requirements. They work fine, but they're just that much harder to get support for -- even here in the forum, the number of people who can give good advice on a VKRC is much less than those who can advise on regular KRCs.
Then there's option packages. Any secondhand robot is probably "stripped" -- you'll want to be very careful about what you buy. A stripped KRC2, for example, only has hardware support for DeviceNet on board, which would be fine, if you stick to DeviceNet I/O hardware (or use an AnyBus converter bridge). Adding a modern bus like ProfiNet or EtherNet/IP to a KRC2 is expensive and painful, and was never fully ironed out before the KRC2s got End-Of-Life'd. Adding a slightly-older bus like ProfiBus is much more solid, but still requires finding the KUKA ProfiBus card, and then finding someone who knows all the quirks of setting it up.
A secondhand robot that's not stripped may have a bunch of special hardware inside that you might be able to use, or might just get in your way.
And most secondhand robots do not come with the original electrical diagrams, manuals, or the OS recovery media. So you want to be very sure what you're getting, and that you can get replacements for all that. If a KRC2 comes off the boat with a dead hard drive, for example, you're probably stuck asking KUKA for a new set of installation CDs to rebuild the system from scratch.
We need to know what kind of accuracy you need and if you are ok with it looking like a beaver chewed on the wood. 
Robots can do some milling/routing but you gotta have a really good setup and know what your doing. No offense but people buying used robots usually aren't the people that will succeed with this type of project.
As Skyfire said "you are NOT going to get the kind of accuracy even a cheap CNC router would achieve".
If that is your requirement you will not succeed. Typical robot accuracy is at least an order of magnitude worse than a CNC.
There are ways to make robots more accurate but it would require buying a new robot and putting extra time and money into it.
