The short answer is that the three are very different. Take note of their differences:
- Cryptography is writing hidden messages – a historical / forensic approach.
- Cryptanalysis is analysing hidden messages – a statistical / analytical approach.
- Cryptology is reading hidden messages – a linguistic / code-breaking approach.
Generally, you’ll see these terms used extremely loosely (if not interchangeably): but that’s something of a tragedy, as each strand is concerned with a different type of discourse, a different type of truth to help us get to the end-line, that of finding out what happened.
(1) If you study the cryptography of the Voynich Manuscript, you would primarily focus on issues such as: the intellectual history behind (and embedded within) the glyphs, the forensic layering of the writing itself, the physical strokes that make up the letters, what corrections there are to be found, how Voynichese practice evolved during the construction of the document, how the writing interrelates with the drawings, etc. This is reconstructive forensic history, that seeks to establish the truth of the writing system – to establish the mental structures that were given systematic shape (and yet were hidden) in the writing. In many ways, the end-product would be an accurate transcription of the text – but I strongly believe that this strand has not yet been pursued to its logical conclusion.
(2) If you study the cryptanalysis of the Voynich manuscript, you would instead take the study of the cryptography completely as a given and use the resulting transcription as a starting point for your analytical research, however (in)accurate it may be. The argument has typically been that even if, say, 10% of the transcription is wrong, statistical analysis of the remaining 90% should still yield informative results that are (to a certain degree) illustrative of the underlying mechanisms. Yet the specific reliance upon the transcription cannot be ignored, particularly when you go hunting for larger-scale patterns (such as words, or lines). And there is a very strong case to be made that the absence of convincing statistical results to date arises not from inadequate statistical testing, but instead from some basic division within the text being misunderstood.
(3) If you study the cryptology of the Voynich Manuscript, then you would take as a given a carefully-selected set of statistical properties previously derived from cryptanalysis, and look for some kind of linguistic fit between those properties and the properties of known languages and/or transformations of known languages (such as shorthand, patois, abbreviation, contraction, etc). Many Voynich theories are based on a very naive cryptological reading, often filling the vast gaps between the two models by expanding the range of possible languages that are present all at the same time, and hence resulting in a claimed plaintext that is a hugely interpretative soup of Romance language fragments – though Leo Levitov’s “polyglot oral tongue” is a prime example, it is very far from being the only one of its kind.