The big problem with these ideas is: How do you make such a complex organism out of DNA assembly? Sure, it's easy to explain the concept, but how do you translate that into the somewhat esoteric DB DNA code? We have to keep in mind that the most advanced-behavior bots currently (that I know of- correct me if I'm wrong) are antbots, tie-movement worms, and moonfisher's NNbots, which are an order of magnitude less complex than symbiotic shell-making multibots.
The biggest single problem I see with this idea, for example, is reproduction- building a wall down the center of the thing is going to be very difficult to program out of the conds, ands, ors, nots, and "is equal to"s of bot DNA. Yeast-like budding would be worse.
I recommend you start with something simple, like A. Minimalis, and try understanding it's code. Then try modifying it to add poison, viruses, etc...
We're here to help if you have any trouble with anything.
Once you've got that down, then you'll probably be able to move onto more advanced projects.
I was a bit intimidated by the DNA code myself when I first saw it, but once you sit down and go through the tutorials it'll just take a couple of hours to master. If you think designing bots is fun, it isn't even close to the fun you get from building them.
Your ideas aren't impossible, of course, but they're just a bit more difficult than what we currently are familiar with. Don't let that discourage you, though. We need creativity here!
Good luck!