That function isn't very graceful, especially with that repetitive shifting and data duplication. Here's how you could rewrite it:

process($data_ref, \@set_up, $template, $fd_out);

sub process
{
     my ($data, $set_up, $template, $fd_out) = @_;

     foreach my $record (@$data)
     {
          # ...
     }
}
[download]

Maybe it's just me, but I find reading function arguments with shift is usually pointless since you can just declare the works in a single line. It's also nice to have the function argument declaration in a similar format to how you call it, so you can see if things match up.

As of Perl 5.6 or thereabouts, you can open a filehandle that is put into a scalar (glob reference) using the open function:

my $fd_out;
open($fd_out, $what_file);

while (<$fd_out>) { # ... }
[download]

Or, of course, there is always IO::File:

my $fd_out = IO::File->new($what_file, 'r');

while (<$fd_out>) { # ... }
[download]

These filehandles are a lot easier to pass back and forth than the regular globs. If you're even thinking about using them as function arguments, don't use globs.

As a note: ${$foo} and $$foo are equivalent. It's often simpler to just leave off the extra braces since they don't serve any practical purpose. There are occasions like ${$foo->{bar}} where they more appropriate.