Actually, almost all popular density functionals are either directly derived from first principles (e.g. PBE, TPSS) or based on a mixture of physically meaningful ideas with parameter fitting to accurate reference data (e.g., the LYP correlation functional or the mixing factors in B3LYP).
I don't want to discourage you, but there has been no major breakthrough in (ground state) density functional theory during the last about 10-20 years. These years have seen many very good ideas (most of them based on first principles) which turned out to not actually work any better than PBE or the B3LYP thing. Recent research has actually mostly focused on patching up holes in DFT, like the dispersion problem, but this is not necessarily done in a terribly elegant way; the main goal is typically to get something which more or less works in practice in 90% of the cases. Progress in fundamental problems like a systematic improvability of functionals or the static correlation problem has been very slow. And there are some reasons to believe that obtaining significant further advances in these areas in the context of pure density functional theory is likely impossible.
Many major DFT gurus have actually given up on that and are now making DFTs which are getting closer and closer to wave function methods... (e.g., optimized effective potential methods, random phase approximation correlation, range-separated hybrids, etc.)
