I've always had the problem with my mixes that a certain instrument or vocal didn't stand out well enough or I needed more separation. So I thought of this (I'll use the example of a guitar muddling out a vocal sample): 1) Take the offending guitar track, copy it to another track, 2) invert it, 3) model the envelope of that track to that of the vocals. Then 4) take a parametric eq to the track and isolate only the freqs that are most common and most essential to the integrity of the vocal track. This would cancel out the guitar freqs that you need the vocals to have, but only when the vocals are present, and only at certain levels. I feel like I'm not wording this very well but you get the point. Comments?