Hi, you can first plot the discrete values you have and then decide for the best model to fit the data (linear, quadratic, .. The difficulty is that you have few data and the model will certainly do not match very well.
Hi, this fits in the problem of bivariate interpolation, a classical matter in numerical analysis. You can get z as a polynomial in x and y. You can also get best approximating functions (linear, Bezier, etc). There are programs that can compute it.
No sorry, I'm not an expert in numerical matters. You may find useful references in such kind of survey : http://link.springer.com/article/10.1023/A:1018981505752
: Did you check the link https://www.mathworks.com/matlabcentral/answers/98925-is-it-possible-to-fit-data-to-more-than-two-independent-variables-using-curve-fitting-toolbox-r2013b