That sounds like a radical change of SFML's current "philosophy". Can you provide some more details of the implementation? Maybe a code example that explains how you'd like to use SFML in the future. (personally I love such examples that don't actually work, but show what features are planned or how the new code should "feel")
Generally I don't think it's a bad idea, you're focussing on the interface that supports things like sprites, texts etc. instead of continuing to develop SFML to a gaming engine -- to be honest, it's already a basic one.
However, I've always liked the fact that SFML enables me to either only access my accelerated hardware or build concrete applications that make use of all the fine little classes that SFML provides.
So I think separating the high-level stuff from SFML is a good idea, but I'd definitely not drop it off from the official release, neither provide the high-level classes as an add-on. What do you think of another SFML package that brings in all the basic high-level drawing classes? Personally I think SFML wouldn't be so useful and popular if it didn't provide those.
Please keep us involved, even in your planning stage. At least I am very interested and also somehow afraid of SFML's future regarding its origin.
(not to mention that SFGUI makes high usage of SFML's high-level classes, so I'd be a bit pissed if you threw them away!
j/k)