No, you should use feature detection, just like you have to do in D3D 11, then it will work as designed on all current and future GPUs. You might need some special cases for drivers that tell porkies about what features they support though.If DX12 is truly a low-level API then you will have to write different code paths for each architecture that you want to optimize for.
If it's Kepler, run this code.
If it's Maxwell, run this code.
If it's GCN 1.1, run this code.
If it's GCN 1.2, run this instead.
This is all very basic stuff here.