Given a graph class $\mathcal{H}$, the task of the $\mathcal{H}$-Square Root problem is to decide, whether an input graph $G$ has a square root $H$ from $\mathcal{H}$. We are interested in the parameterized complexity of the problem for classes $\mathcal{H}$ that are composed by the graphs at vertex deletion distance at most $k$ from graphs of maximum degree at most one, that is, we are looking for a square root $H$ such that there is a modulator $S$ of size $k$ such that $H-S$ is the disjoint union of isolated vertices and disjoint edges. We show that different variants of the problems with constraints on the number of isolated vertices and edges in $H-S$ are FPT when parameterized by $k$ by demonstrating algorithms with running time $2^{2^{O(k)}}\cdot n^{O(1)}$. We further show that the running time of our algorithms is asymptotically optimal and it is unlikely that the double-exponential dependence on $k$ could be avoided. In particular, we prove that the VC-$k$ Root problem, that asks whether an input graph has a square root with vertex cover of size at most $k$, cannot be solved in time $2^{2^{o(k)}}\cdot n^{O(1)}$ unless Exponential Time Hypothesis fails. Moreover, we point out that VC-$k$ Root parameterized by $k$ does not admit a subexponential kernel unless $P=NP$.