ROCR is a good R package for computing ROC-AUC values and plotting many types of curves such as ROC, precision-recall, lift charts and so on. It supports cross-validation in the sense that it can plot average ROC (and other) curves for multiple runs with error bars in a way that you don't need to calculate averages and standard errors yourself. The cross-validation itself is just a matter of writing a loop where you select a different subset of the data at each iteration.

You could use Weka, a popular general-purpose data mining software. It supports a variety of attribute selection schemes that you could use out-of-the-box, and some can be configured to use crossvalidation AUC as a guiding criterion. For instance, you could use the "WrapperSubsetEval" which can use any of the supported classifiers in Weka (and there are many!) to find a subset of features that maximizes a classifier's AUC.

Or, you may want to check out Weka's SVMAttributeEval which iteratively removes attributes by using attribute weights derived from Support Vector Machine models (not AUC!) and has been shown to work well on microarray data (see Guyon I et al, Machine Learning 2002, 46:389-422.)

I use caret R package, it's flexible and quite easy. It has built-in cross-validation and other tools. You can also train for optimizing AUC.