Suppose a classical Hamiltonian system with canonical coordinates. A function has associated a Hamiltonian vector field with a flow. Now, we can define a family of transformations in the algebra of smooth functions (i.e., observables)
by means of
This is completely related to the Heisenberg vs Schrodinger picture discussion. The flow in is some kind of evolution of states, like in the Schrodinger picture. And the flow corresponds to Heisenberg picture.
Observe that
and
And also, respect the Poisson algebra structure.
Reciprocally, if we start with a function and we are able to associate a family of transformations from to that preserve the structure of Poisson algebra and such that
we can define a flow in such that
We proceed by taking the vector field and the associated flow , and using the two required properties, above. Since
and
we can show that by fixing a point .
We indeed have a duality, between transformations in and transformation in . Why do we pass from the flow in the phase space to the "flow of observables"? Because this way, I guess, we can formulate everything in terms of algebras, and we get a deeper insight in the connection with Quantum Mechanics. Compare this with the discussion of derivations and 1-parameter group of automorphisms of the matrix algebra.