Well the simplest answer to this is that when we carry out such a process, it works. Random variation plus selection produces functional objects, indeed objects of a sophistication greater than we could deliberately design.
We can also of course observe it working on a small scale on organisms.
So unless you can think of some theoretical reason why it shouldn't work in the particular case when you don't want it to, then the logical conclusion is that it does.
Also to return to the topic of this thread we have the transitional fossils, which do make it look awfully like it actually happened.
I did spell out some time ago the differences that have to be navigated to get from the reptilian ear to the mammalian ear, which is a much more limited project than getting from a cow to a fish but I may try that one two. Remove legs from genome. Add fins and fish tail and fish breathing apparatus. This is silly. Just to make those two changes would require millions of years of mutations and still you wouldn't get them.
Fish aren't actually descended from cows, so your argument (such as it is) is about as pertinent as when Buzsaw, God rest him, proved that the Mississippi River couldn't possibly have produced the Grand Canyon.
It really doesn't matter what you want to descend from what. It's impossible unless the genetic stuff is already present for the completely new creature. For instance getting a human being from a chimpanzee is equally impossible. Each has its own genome that makes the given creature and nothing else. To get from the chimpanzee to the human being requires a whole slew of changes into something altogether different from the chimpanzee, things that don't exist in the chimpanzee genome and that makes any such transformation impossible because mutation has to invent something completely new rather than simply changing something that already exists into another version of it which is already potential in the genome anyway.
(1) How are you distinguishing between things which are "another version" and things which are "completely new"?
(2) How would we distinguish between these two things on the chromosomal level? I.e. if I showed you two sequences of A G C and T, how would we determine if one was something "completely new" compared to the other, or if it was just "another version" of it?
(3) Why do you claim that the production of something "completely new" is impossible? You have supplied no evidence nor argument but just assumed this as an axiom.
Randomn variation plus selection gets you a version of the same creature, it's nothing but microevolution or standard variation. You can't get macroevolution out of any such events. The reason it doesn't work is that macroevolution requires something completely new to be formed. All those mechanisms do is rearrange what already exists.
The last sentence is of course complete bullshit, as you would know if you knew anything at all about genetics. This invalidates the nonsense that precedes it.