Nope. Don't do this.
* Design your denormalizers so they are replay friendly. Thats one of the big perks of CQRS is the ability to replay the events and generate new or modified read models. Having your event handlers issue commands (that generate more events) removes this characteristic.
There is another way which is possible using Ncqrs.Messaging that allows your aggregates to dispatch "Messages" (that act like commands) to other aggregates.
I had much trouble with this approach and ended up abandoning the entire concept of "ncqrs aggregates" issuing command like messages.
Instead I switched over to NServiceBus based Sagas/Process Managers that act as my Bounded Context and Aggregate messaging orchestrators.
For example I have a process manager called: PendingOrderApprovalProcess
This saga listens for EVENTS generated by my domain and handles the construction ,ordering and dispatching of related commands to fulfil the process of approving an order.
Even tho the Saga is an event handler, you should have this in a seperate concept seperate to your read model denormalizers. (Obviously replaying events should not re-run the PendingOrderApprovalProcess).
I hope this helps.