I would rewrite it as a first order system,

v'=sqrt(f-v-y); y' = v

and solve it numerically, keeping the two possible signs of the square root in mind.

I don't think there is much hope for analytic solutions, except perhaps for very special cases of f(x).