Alternative output layer after CAFFE model import

I would like to be able to use the CAFFE model OpenPose in the Deep Learning Toolbox. The network as available has as an output layer a concatenation layer, which produces a set of (height x width x N) images, which are probability maps for various body joints or linkages between those joints. Matlab will not accept that as an output layer, so I have been adding an arbitrary output layer into the importCAFFEmodel command, and then editing the layers in deepnetworkdesigner. I need to replace the output layer, but I am not sure what to replace it with. I need something like a classification layer that will work on multi-layer set of images. Has anyone tackled this already, or have suggestions on what a custom output layer woudl look like in this situation? Thanks.