I am very basic as things pertain to statistical software in the sense that I have been sticking to SAS since the 1980's. The code stayed the same all these years. and when we started using SAS on PC's, the only addition to the code was a last line with RUN; Not many software packages can survive for so long, and not many packages are as complete as SAS is.
I use SAS in all of my statistical analysis, except for special applications, such as cluster analysis, where I use SaTScan.
I'm afraid that is all I can say here about a software for nonparametric regression.
Recently, applications of support vector machine algorithms have demonstrated a sensitive improvement in results compared to that obtained with PLS, especially for data with high nonlinear relationships or complexities. Artificial neural networks has also been used for nonlinear multivariated calibration, however the final results depend on the initial parameters, sometimes is necessary to repeat the net-work training, the final solution is non-unique and it has the tendency to overfit. SVR has the advantage in relation to ANN in produce a global model that capable of efficiently dealing with non-linear relationships.
I second the recommendation of R. Just find the right package, spend about a day playing around with the code, read the package documentation, and you're good to go. It's fairly straightforward to run Kendall tests and Theil-Sen estimators.
If you are willing to spend a little time learning it, Python, with the SciPy stack, is easily the best language for any mathematics or statistics (and it is open source and free). The package you would want is here: https://pypi.python.org/pypi/PyQt-Fit
and he uses R and S-Plus - providing many examples with data and code.
If the data structure is more complex - eg involving space and time - then this free software has excellent facilities for fitting generalized additive models:
It"BayesX is a software tool for estimating structured additive regression models. Structured additive regression embraces several well-known regression models such as generalized additive models (GAM), generalized additive mixed models (GAMM), generalized geoadditive mixed models (GGAMM), dynamic models, varying coefficient models (VCM), and geographically weighted regression within a unifying framework. Besides exponential family regression, BayesX also supports non-standard regression situations such as regression for categorical responses, hazard regression for continuous survival times, and continuous time multi-state models (see model classes and model terms below). "
see http://www.stat.uni-muenchen.de/~bayesx/bayesxwhat.html
I find it interesting that SAS is not being recommended for nonparametric regression. Is the main reason the cost or do you prefer other software packages for better and more advanced algorthms?
R is a good option , for sure. But I think STATA is the easiest software to perform a non parametric regression (Quantile regression > median regression) . The slope b of the regression (Y=bX+a) is calculated as the median of the gradients from all possible pairwise contrasts of your data. A confidence interval based upon Kendall's t is constructed for the slope.