You can probably do this using an undocumented(?) feature of MathExpression. The expression language used in MathExpression (and AddExpression) supports calling static Java methods as part of the expression (which actually makes the NumericTransform filter in WEKA redundant). For example, to add uniform noise +/- 0.5 to your numeric target, assuming it is the last attribute in your data, you can use
weka.filters.unsupervised.attribute.MathExpression -E "A-0.5+java(\'java.lang.Math\', \'double random()\')" -V -R last -unset-class-temporarily
Note that the expression language also supports conditional statements, so you can do pretty sophisticated things with it.