As a general technique, here's a way to emulate  [^set] where you are dealing with a potentially complex regex expression rather than a simple character class:

c:\@Work\Perl\monks>perl -wMstrict -le
"my $cd     = qr{ cd }xms;
 my $not_cd = qr{ (?! $cd) . }xms;
 ;;
 for my $s (',abcdefg,pqrstuv', ',abefg,pqrstuv', @ARGV) {
   my $t = $s;
   print qq{'$t'};
   $t =~ s{ , $not_cd* ($cd?) [^,]* , }{=$1=}xms;
   print qq{'$t' \n};
   }
"
',abcdefg,pqrstuv'
'=cd=pqrstuv'

',abefg,pqrstuv'
'==pqrstuv'
[download]

Where:
  $cd is an arbitrary regex pattern;
  $not_cd matches any character that does not begin the arbitrary pattern.

Update: Here's an example more in tune with your OPed assertion that the start and end delimiter patterns (and the optional included pattern) may be complex:

c:\@Work\Perl>perl -wMstrict -le
"my $maybe = qr{ cd? }xms;
 ;;
 my $start    = qr{ A | BC | DE?F }xms;
 my $end      = qr{ U | VW | XY?Z }xms;
 my $excluded = qr{ (?! $end | $maybe) . }xms;
 ;;
 for my $s ('AxxcdxxVWABCDEFUVWXYZ', 'BCxxxUABCDEFUVWXYZ', @ARGV) {
   my $t = $s;
   print qq{'$t'};
   $t =~ s{ $start $excluded* ($maybe?) .*? $end }{=$1=}xms;
   print qq{'$t' \n};
   }
"  BCxxcxxXZABCDEFUVWXYZ
'AxxcdxxVWABCDEFUVWXYZ'
'=cd=ABCDEFUVWXYZ'

'BCxxxUABCDEFUVWXYZ'
'==ABCDEFUVWXYZ'

'BCxxcxxXZABCDEFUVWXYZ'
'=c=ABCDEFUVWXYZ'
[download]

The  .*? $end could be  $excluded* $end instead. This would make the regex perhaps a bit more robust, but a bit slower.