The theory should also be flexible, and change as you get to more and more obscure things. And take into consideration the ecosystem of what actually uses that feature.
V4L, almost by definition, deals with unusual and possibly obscure hardware. The usecases might neatly fall into: stable systems with old hardware that never changes, is never upgraded, etc. And: bleeding edge systems with new and cool hardware, tweaked to the Nth degree by some geek giant.
At the very least: V4L deals with hardware. Old hardware? Old kernel. New hardware? New kernel. That seems fair and reasonable to me.