I need a statement that must be classical but it's being very difficult (even after consulting experts) to find a reference. Let's try here!
Let $f$ be a function on the boundary of a bounded domain $\Omega$ in $\mathbb R^n$ and let $w$ be the single layer potential. Namely, $w(x)=\int_{\partial\Omega} f(y) P(x-y) dH^{n-1}$, where $P$ is the Newtonian potential.
If the domain is $C^{k, \alpha}$ then it should be true...
$f\in C^{k-1,\alpha}$ $\Rightarrow$ $w\in C^{k,\alpha}(\overline{\Omega})\cap C^{k,\alpha}(\overline{\Omega^c}) $ ($k+\alpha$ derivatives up to the boundary for each side)
$f\in C^{k-2,\alpha}$ $\Rightarrow$ $\frac 1 2 (\partial_{\nu, {\rm in}}+ \partial_{\nu, {\rm out}}) w \in C^{k-1,\alpha}(\partial\Omega)$, where $\partial_{\nu, {\rm in}}$, $\partial_{\nu, {\rm out}}$ denote the normal derivatives form inside an outside. (we should obtain an extra gain do to the typical cancelation...)
This type of result in for $k=1$ is contained in classical books of potential theory and PDE from math physics (Sobolev, Miranda, ...). I would like to find (if it exists) a clean modern reference for all $k$. It is possible that one needs to flatten de boundary and then apply a general enough result (which I do not know) on boundedness of Riesz-type transforms...