In the course of developing a system, the analyst's understanding of the system will change in light of the analysis they do. The same is probably true of the customer.
So in the course of analyzing a system, the analyst will often make several versions of each type of model either building upon earlier ones or even throwing old ones away and starting again.
It makes sense then to have models which are reasonably cheap and can be developed and maintained without too much cost and time overhead.
For this reason, the models often take the form of diagrams, pictures, and text which can be printed on paper or drawn on a whiteboard.
This makes them very cheap! Occasionally, more sophisticated models are used, such as software prototypes in the prototyping life-cycle model. The models used should also be easy to change.
Re-writing or re-drawing models by hand is tedious, and CASE software tools can be used to relieve the burden of producing models.
More fundamentally though, the models themselves should be such that adding or removing detail does not have many repercussions on other, distant parts of the model. In most cases, this means that the model should not represent the same piece of information in two different places.
This property of reducing multiple representations of the same thing is called minimal redundancy.
The property of changes to a part of a model having minimal repercussions on other parts is called low viscosity or robustness to changes. A good model will usually have both minimal redundancies and be robust to changes.