If you have a fixed set of symbols that you use on a block of text, those symbols are still going to be used in whatever frequency is typical for the use frequency for their substitutions for that language. What he needs to do is to have the letter substitution be changed according to some predetermined standard such as the first 100 letters of the message are encoded by the novel alphabet where A to Z = 1 to 26, the second 100 are encoded where A to Z = 5 to 26 and back through 4, etc, varying both the numbers in succeeding groups and the starting point in the novel alphabet for each. If one doesn’t know the varying group length or the start point for each group in the novel alphabet, one can’t easily look for any kind of frequency distribution for particular letters such as “e” or for common letter combinations such as “ed” “ing” “ly” and double letters such as “tt” in “letter,” “bb” in “bubble,” “dd” in “added,” etc.
Yeah, but your premise is useless. He invented a new language that only he and the second party knows. The way to crack this code isn’t to crack this by frequency of any letters or letter combinations alone, but rather comparing character frequency to all known languages, finding out what family of languages this constructed language is based on and decipher some of the cognates and frequently used words in that constructed language.
It won’t be a perfect cracked code, but you may be able to pick off a few simple words via cognates and frequently used words. All constructed languages are related to some language family or the other.