I found this doc. on the web somewhere. The main problem with building your own NMR is the magnet. Commercial models use a superconducting magnet to get the required field strength and uniformity. Recently, there has been some progress towards achieving the required magnetic field using multiple rare earth magnets in a Halbach arrangement. The use of commercial RF amplifiers to drive the coils and offloading signal processing to the PC means that a desktop NMR spectrometer is within sight although not yet achieved. Building one would not be easy and costs would be in the order of a couple of thousand euros. Amongst other stuff on the web that can be googled, this report is quite interesting.
http://rapidshare.de/files/1614980/nmr.doc.html