Now, this is according to the knowledge I have leanred throughtout the months and may not be entirely accurate...
A co-processor is also more specifically and properly known as a math co-processor, and is used for the single purpose of computing math and number information.
Normal processors of the early computer era could not process direct mathematical and numerical equations and information because of their archetechture; it just couldn't support it. The only way it could process complex math functions was to be rendered first down into binary language from a compiler and then processed, an action and process that is quite time extensive. What a math co-processor was quite basically was a calculator processor, seperate of the CPU, that could drastically cut down the processing time of certain functions and programs because that was it's only dedictaed purpose, while the CPU could go on processing what it normally did at its usual speed.
Programs with 2-D/3-D graphics, like AutoCAD which also uses an immense amount of mathematical equations and measurements to produce the drawings, therefore can utilize the conveinance of a math co-processor to not only cut down the processing time by half, or even more, but keep the processing load off of the main processor/CPU, therefore extending it's service life.
So you -could- use a program like AutoCAD without a co-processor installed in your system, but it would be so rediculously slow to render any drawings and measurments as to be nearly useless. Installing a math co-processor would pretty much be a nessecity to be productive.
When the 486 CPU came out from Intel in 1989, it was the first CPU to have a built-in math co-processor, therefore eliminating the need for external and seperate co-processors because software at that point in history was getting advanced enough to warrant this mergance. This is why you see no math co-processors for any 486 processor or other CPU made after 1989.