How do you expect to use the resulting correlation; as function which variables?
You use nonliner fitting; convert the measured data to values of the correlation at the points for which you have the values at those points for which you have the data.
You may have issues of correlation/estimation depending on the locations of those points; you've not given that much info; try it and find out what happens...
As for which tools to use for the fitting, depends on which toolboxes you have; there are routines in either the Curve Fitting or the Statistics TBs or you can use the fminsearch routine in base product having neither of the above....
