Programming :: How To Implement Whole New Writing System

So basically, there is a really cool writing system I have been working on. It could be viewed (for simplification purposes) like an encryption method for the Latin script.

Facts about the writing system:
It has a little over 300 symbols.
It is syllable-driven.
It is highly compositional (eg. "c", "ca", "cae", "ca " and "ci." all map to different symbols - and NOT by overlaying elements)
Symbols have medium graphical complexity (comparable to Korean Hangul, or Japanese Hiragana)
Has a rather complex set of diacritics (~10, some of which can go on any symbol)
Has no ligatures

How transliteration occurs:
Sequences of Latin symbols map to certain symbols. Example below:[G][rou][p ][hu][g.]Characters sequences between "[" and "]" map to a single symbol (so it would take only 5 symbols to write "Group hug.").

How I want it to work:
I would like to have a daemon that: Intercepts all text displayed on the screen.
Converts it to my writing system (changes letter sequences with individual Unicode codes)
Leaves unsupported symbols unchanged.
Displays all the text on the screen using my font and characters intertwined with the fonts and characters left unchanged.

For example, if you take the following line of C++ code:for (i = 1; i <= n; i++)
I would like it displayed like this:[fo][r ]([i ]= [1]; [i ]<= [n]; [i]++)
Bold-symbols should be in my Unicode font with special symbols defined for this writing system, and the rest should be in its original font and encoding. Also, I would like this encoding to hold for display-purposes only. The data in the memory should remain unaffected. This also means real-time adjustments: if I open a text editor (say, from the OpenOffice Suite) and I start typing, I would like to see what I type encoded with my writing system, even though the document actually contains Latin letters. This also means that the symbol immediately before the cursor may change as you type.