Another way of thinking about this is: Compress with one Huffman compression scheme based on the message text then Decompress with the Huffman scheme based on the desired statistics.