Sounds like you just need eval. I suggest that you turn your rules into subroutines that can be cached.

#!/usr/bin/perl -w 

use strict;
use warnings;

# Build list of translation rules
my @rules;
while (<DATA>) {
    chomp;
    my ($pattern, $code) = split "\t";
    push @rules, [$pattern, eval "sub {$code}"];
    warn $@ if $@;
};

# Fake Data
my @lines = (
    "this is foo\n",
    "this is bar\n",
    "this is baz and foo\n",
);

# Process the lines using compiled rules.
for my $line (@lines) {
    for my $rule (@rules) {
        if ($line =~ /$rule->[0]/) {
            $rule->[1]->($line);
        }
    }
    print "$line";
}

__DATA__
foo    $_[0] =~ s/foo/FOO/;
bar    $_[0] =~ s/bar/BAR/;
baz    $_[0] =~ s/baz/BAZ/;
[download]

This way you can use whatever perl translations that you want. Also, it would be up to you what you pass to the subroutines and if it needed to return anything.