Philosophical answer: You should separate the model from its representation. → MVC

Assuming, your CGI returns HTML, it is not a good idea to scrape the information back. Let your code call the sub for data and let the sub return the data only. Let your CGI-sub call the same sub and let it render the result in a specific format that is returned via HTTP.

IF your CGI returns data in a format that is parsable (i.e. JSON or YAML or XML or ...), then there's the alternative to call the CGI via HTTP (LWP::UserAgent) and deserialise the result. That's not too performant, though and requires additional error-handling. On the pro-side, data can be retrieved from a remote server.

Hope that helps. I might completely misunderstood your question, though ;-)

Update: Clarification in form of code fragments:

#-- the sub that computes the data (@result) only from it's parameters
+ (@pars)
#   Using @result as an array is just for simplicity. It could be hash
+-ref or an object
#   or whatever your expected result fits into...
#
sub get_result { my (@pars) = @_; ..... return @result; }


#-- here's your CGI handler
#
sub cgi_handler {
                   ...
                   my (@pars) = ... #-- get params from HTTP request
                   my (@result) = get_result( @pars ); 
                   ...              #-- render HTTP-response from your
+ @result
}


#-- somewhere else in your code... where your original question came f
+rom...
#   Instead of a regular function, an object method or a class method 
+or an AUTOSUB etc. 
#   could be called.
#
sub somewhere_else {
  #-- no worry to call cgi_handler(), just call
  @result = get_result( @pars );
  ...
}
[download]