Even the pros have a hard time with this stuff, guys . . . it's impossible to fully separate elements that have been mixed together into one track. Not that there aren't ways of reducing them, depending how the mixing was done, but it ain't easy. I admit I don't know a whole lot about this stuff myself.

Just yesterday I had the opportunity to hear a Neve stereo processor intended for mastering use, which among other things is able to perform some very sophisticated mid/side processing techniques to affect stereo imaging. By splitting the signal into central information and wide information and treating those independently, many tricks and alterations are possible when recombining them into left/right. This thing can definitely reduce the level of the vocals drastically, which impressed me a whole lot, though it isn't a perfect separation by any means. Also, it isn't possible to distinguish between vocals and centrally located elements of the music (though some judicious use of EQ could help somewhat), so what's left over isn't going to be a complete version of itself.

If you really want to try your hand at this kind of thing, find yourself a good mid/side processor of some kind and learn how to use it; that's my advice. But be aware that they're really intended to be used to improve existing mixes by clarifying stereo imaging and things of that nature, so don't expect miracles of separation.