If the font data isn't already compressed (I'm guessing not from the description) then a simple decompression algorithm after base-64 decoding might reduce the

If the font data isn't already compressed (I'm guessing not from the description) then a simple decompression algorithm after base-64 decoding might reduce the size considerably more, although there is a direct trade-off against added code size, given that it's a one-off. Note that I say "simple decompression" (not compression) because the compression itself can be arbitrarily complex, to optimise the compression ratio. Such asymmetries are not uncommon in compression algorithms, so it probably wouldn't be too hard to find one that fits the bill here.