Estimating the non- parametric conditional probability

Question

I have a set of observation from two parameters, let say $x$ and $y$ and then I want to make the conditional probability of $x$ for the given $y$, $p(x|y)$. So first I use scipy.stats.gaussian_kde to obtain the joint probability distribution of $p(x,y)$ and use normalized histogram and one dimensional interpolation to obtain $p(y)$ and then in the grid points of $x$ and $y$ plane, I compute $\frac{p(x,y)}{p(y)}$ in order to get $p(x|y)$. I also want to measure cumulative distribution of the conditional probability in the grids. Is it the correct approach to estimate the conditional probability or there are other efficient and faster approaches?