For a similar issue, Keith suggested the following approach:

1. join paired reads into single megareads (using Galaxy toolkit)
2. do your thing (filtering or barcode splitting)
3. split the concatenated megareads back into paired reads

cheers,
y.

I had the same problem with filtering paired-end reads. In the end, I just wrote a script to call the fastx-toolkit filter on both files and then further filter out reads that were removed from only one of the pairs. Since you know the files come out in the same order, you can do the filtering yourself without reading everything into memory.

So, yes, a custom script is likely your best bet as most existing tools only do a single-end at a time.

NovoBarCode handles paired end Illumina reads and works with a number of different barcoding schemes:

http://www.novocraft.com/userfiles/file/NovoBarcode.pdf

It's not open source, but is free for academic and non-profit use.