can anyone please help with Bayesian estimatiors?

If my model is a function of unknown parameters I am drawing inference about, say Weibull distribution with scale s and shape m:

W(m,s),

what is the proper Bayesian estimator for that model?

I see two options:

1) evaluate the model for point estimates (e.g. a posterior mean) of s,m

W(E(m), E(s))

or

2) evaluate the expectation of the model as a multivariate function of random variables

E(W(m,s))

The second option seems correct from a math point of view but has some practical disadvantages. The properties of the original model are lost, i.e. the result is not a Weibull distribution anymore.

Any help or hint to literature are very appreciated!