Subclassing numpy array for symmetric matrices

I want to create a numpy subclass specialized for symmetric matrices.I want it to be just like any numpy array with only 2 differences:

1. It will have 2 fields for its eigendecomposition: U, D.
2. On construction it will force symmetry.
   Like my_symmetric_arr() = 0.5*(m+m.T) where m is the regular matrix created by the default constructor.

How can that be done in the simplest way?