The code you've shown does not compile, and is hard to read. Please take a look at SSCCE, perltidy, and Use strict and warnings.

When I compare $a and $b, then I want to catch only "A". Also When I compare $b and $c, then I want to catch "B".

But that's not really what your code is doing - it's bunching all the characters from all the strings together into one hash, and then looking for unique characters. Could you explain your requirements, along with more examples?

but there are so many loop that script's performance is really bad

When asking questions about performance, it's useful to provide as much relevant information as possible. How big is your input data? What does "really bad" mean - how long does it take to run? And how quickly do you need it to run? Is the code you're showing really the code you're running? Again, please provide an SSCCE that is representative of what you're actually doing.

Until you give more details, I'll just give one hint: If you have two strings of the same length, you can do an XOR operation, as in $x^$y, and the resulting string will have NUL bytes ("\0") wherever the strings were identical, and non-NUL bytes where they differ. But again, if this is "better" depends on your actual requirements.