I've not used this myself, but consider using the NFFT posted on Chemnitz University of Technology Department of Mathematics site. It reduces the O(N^2) requirement to just O(NlogN) as in the FFT case. Also, it now includes Matlab classes to interface with the mex files.
You can download some examples on the site that show to to interact with MATLAB and the faq has instructions on how to use in Windows+MATLAB (if you are).
NFFT requires the initialization of a plan and precomputes several things to increase performance. It looks like it will take some effort to get familiar with but may be very helpful to you.
It is licensed under the GPL.