The problem with stealth in RPGs is that to do it right or well, you have to abstract the conflict. A lot of games (and a lot of players) will fight you to the death to avoid abstraction as presentation, claiming that it destroys their immersion, is too "game-y", and they'll do it while they bitch about how stealth just can't be modeled well and "why isn't there a guard over there?"
The problem for very traditional games like D&D is that conflicts within the mechanics are typically between individuals, and stealth just is not a test between individuals when you're talking about sneaking. Stealth is a greater environmental state which the player or players are attempting to overcome.
The best way to represent that in most games is to simply require X successes within Y time to consider the intent of getting through the stealth section achieved. If you get more successes within that time, keep those in your pocket and use them to do something cool because the enemy didn't see you coming. If you get fewer successes within that time, you have not approached stealthily – and if the group/GM is feeling particularly evil, they'll pocket the difference in successes to use against the protagonists during the next steps because the enemy heard them/saw them/detected them coming.
The best part about this abstracted mechanic is that it works to allow a test for any sort of preparation/buildup/carefulness at the beginning of a particular set of conflicts or right in the middle. You can even keep it going, requiring – as an example – two points of success per turn for every party member attempting to remain stealthy. Sure, you are Ninja finds it no problem at all to cover for himself and maybe even one other person to split away from the group, but they are going to have to keep it quiet out there to avoid being detected, and he's going to have to bank up some stealth successes along the way to help cover the turns in which combat are occurring.
I've never seen all of these mechanics put together in the same place, but it would be easy enough to do. Wushu is the obvious inspiration for this, really, because it already uses invested success pools to talk about long-term plans and preparation. But any system in which you have degrees of success or numbers of successes would be perfectly reasonable to bolt this on.