This seems rare in practice. I've only seen 魔 written as ⿸广マ
Character Decomposition
The vast majority of Chinese characters are composed of radicals and other characters. There have been attempts to systematically define characters by their components, but none seem to have stuck around.
Defining characters by their components would solve issues related to CJK unification, in which identical Unicode code points are rendered differently depending on linguistic context. (澀/涩/渋/澁 is an example of a character that varies greatly by region)
Cangjie Input constructs characters from basic components designed for QWERTY Keyboards. Because it is an input method, it only outputs Unicode Chinese characters.
Konjyaku Mojikyo is software and font which stores and renders character structure
It even supports Tangut!
Recursive Radical Packing Language (RRPL) is a schema and dataset for defining characters completely from stroke data. Stroke definition can (recursively) reference radicals and other characters. The main purpose is for machine learning. It can handle arbitrary characters.
Unicode includes Ideographic Description Characters (U+2FF0 to U+2FFF) which notate character structure, but they are not used for rendering "illegal" characters.
Duang (⿱成龍) is a Chinese internet meme originating from a commercial featuring Jackie Chan. The word describes the sound his hair makes (?) and is written by combining the characters in his Chinese name vertically. Because you can't type the character, people insert an image into text. The reading is also "illegal" in Mandarin - no Chinese word is read "duang".