Ah indeed, AxB lets you do only 1xN across separate files, so not a solution for you.
Then I don't think there is any solution other than computing the full NxN matrix and keeping only the values that are interesting for you.
If you really don't want to see the other values, you can explicitly set them to zero in the matrix. You'd need to edit the TF field in the connectivity file manually: https://neuroimage.usc.edu/brainstorm/Tutorials/Scripting#Custom_processing